Current hypothesis is that the socket timeouts added in #415 have been created additional problems: the default socket timeout is "infinite", and I've just added the read timeouts to the place where they weren't supposed to be.
I've tested the slightly changed horta version (commit chain ac08ff1f39f59cc53678f40a130b6a5b90e0006f → f0aeb1870cbceb14eb4fd149082e4aac603aaeb1), and it was reconnecting every 2 hours; probably that's just some XMPP stuff, and definitely not the problem. I'll file another issue about that (maybe we just need to tune XMPP ping timeouts a bit).
I don't think that commit ac08ff1f39f59cc53678f40a130b6a5b90e0006f was essential as a solution to the issue, and that's the reason I've removed that commit from the PR (it will arrive later as part of #408).
This is another try to fix #414.
Current hypothesis is that the socket timeouts added in #415 have been created additional problems: the default socket timeout is "infinite", and I've just added the read timeouts to the place where they weren't supposed to be.
I've tested the slightly changed horta version (commit chain ac08ff1f39f59cc53678f40a130b6a5b90e0006f → f0aeb1870cbceb14eb4fd149082e4aac603aaeb1), and it was reconnecting every 2 hours; probably that's just some XMPP stuff, and definitely not the problem. I'll file another issue about that (maybe we just need to tune XMPP ping timeouts a bit).
I don't think that commit ac08ff1f39f59cc53678f40a130b6a5b90e0006f was essential as a solution to the issue, and that's the reason I've removed that commit from the PR (it will arrive later as part of #408).